A multi-orbital iterated perturbation theory for model Hamiltonians and real

material-specific calculations of correlated systems

Nagamalleswararao Dasari,∗ Wasim Raja Mondal, and N. S. Vidhyadhiraja†
Theoretical Sciences Unit,
Jawaharlal Nehru Centre for Advanced Scientific Research,
Jakkur, Bangalore 560 064, India.

Peng Zhang, Juana Moreno, and Mark Jarrell

Department of Physics & Astronomy and
Center for Computation & Technology,
Louisiana State University,
Baton Rouge, LA 70803-4001, USA.
arXiv:1504.04097v2 [cond-mat.str-el] 2 Feb 2016

Perturbative schemes utilizing a spectral moment expansion are well known and extensively used
for investigating the physics of model Hamiltonians and real material systems (in combination with
density functional theory). However, such methods are not always reliable in various parameter
regimes such as in the proximity of phase transitions or for strong couplings. Nevertheless, the
advantages they offer, in terms of being computationally inexpensive, with real frequency output at
zero and finite temperatures, compensate for their deficiencies and offer a quick, qualitative analysis
of the system behavior. In this work, we have developed such a method, that can be classified
as a multi-orbital iterative perturbation theory (MO-IPT) to study N-fold degenerate and non
degenerate Anderson impurity models. As applications of the solver, we have combined the method
with dynamical mean field theory to explore lattice models like the single orbital Hubbard model,
covalent band insulator and the multi-orbital Hubbard model for density-density type interactions in
different parameter regimes. The Hund’s coupling effects in case of multiple orbitals is also studied.
The limitations and quality of results are gauged through extensive comparison with data from the
numerically exact continuous time quantum Monte Carlo method (CTQMC). In the case of the
single orbital Hubbard model, covalent band insulators and non degenerate multi-orbital Hubbard
models, we obtained an excellent agreement between the Matsubara self-energies of MO-IPT and
CTQMC. But for the degenerate multi-orbital Hubbard model, we observe that the agreement with
CTQMC results gets better as we move away from particle-hole symmetry. We have integrated
MO-IPT with density functional theory based electronic structure methods to study real material
systems. As a test case, we have studied the classic, strongly correlated electronic material, SrVO3 .
A comparison of density of states and photo emission spectrum (PES) with results obtained from
different impurity solvers and experiments yields good agreement.

PACS numbers: 71.27.+a, 71.10.Fd, 71.10.-w, 71.30.+h, 71.15.Mb

I. Introduction (LMA), non-crossing approximation (NCA) and fluctua-

The development of efficient methods to solve quan-
tum impurity problems, especially those involving multi-
ple orbitals, has been a significant research direction in
the field of theoretical condensed matter physics. Sub-
sequent to the development of the dynamical mean field
theory (DMFT)1 , which is exact in the limit of infinite
dimensions and an excellent local approximation in fi-
nite dimensions, the importance of obtaining reliable so-
lutions to general quantum impurity problems has in-
creased further.
Within the DMFT framework, a lattice model may
be mapped onto a quantum impurity embedded in a
self-consistently determined host. The impurity problem
may then be solved by a variety of techniques including–
numerically exact methods like quantum Monte Carlo
(QMC), numerical renormalization group (NRG), exact
diagonalization (ED) and density matrix renormaliza-
tion group (DMRG) or semi-analytical methods like iter-
ated perturbation theory (IPT), local moment approach
tion exchange approximation (FLEX). Each method has
its own advantages as well as pitfalls. For example,
QMC2 is a numerically exact method, but is computa-
tionally expensive. It yields data on the Matsubara axis
(or imaginary time) so to obtain dynamical quantities
such as the density of states and transport quantities,
analytic continuation of the data to real frequencies is
essential3 , which is a mathematically ill-posed problem.
Additionally, it is very difficult to access the low temper-
ature region where statistical errors become important.
As a real frequency method, NRG4 can avoid the difficul-
ties that arise from the need to carry out analytic contin-
uation. However, the method becomes extremely cum-
bersome for more than one impurity or channel. NRG
is better suited for low temperature studies. Recently,5
NRG was applied to study degenerate multi-orbital lat-
tice problems, but the non-degenerate case remains un-
explored. ED6 is also a real-frequency method, but one
considers only a finite number of bath states, so the re-
sulting energy spectrum is discrete, and the broadening

procedure for obtaining continuous spectra is not free of
ambiguities. Moreover, large systems or multi-orbitals
are not accessible. DMRG7 for the single site case has
some numerical artifacts and its accuracy as an impurity
solver is not entirely clear2 .
The semi-analytical methods are perturbation the-
ory based solvers that attempt to capture the essential
physics by constructing an ansatz for the single-particle
quantities. The ansatz is based on satisfying various lim-
its or conservation laws, and comprises diagrams up to
a certain order or sums a specific class of processes to
infinite order. The main advantages of these methods
are that they are computationally less expensive than
the numerically exact methods listed above, while also
yielding real frequency data. However, semi-analytical
methods are, by definition, approximate and need to be
benchmarked against exact results to gauge their range of
validity. For example, although NCA8 gives qualitatively
correct results for temperatures higher than the Kondo
temperature, spurious non-analyticity at the Fermi en-
ergy develops at lower temperatures9 . To recover the
correct Fermi liquid behavior at low temperatures, one
needs to consider a larger class of diagrams10 . The FLEX
approximation is conserving in the Baym-Kadanoff sense,
but it does not have the correct strong coupling behav-
ior. So when it is employed for the half-filled Hubbard
model, strong coupling physics like the Mott transition
is not captured11,12 . The FLEX13 has been extended
to study degenerate multi-orbital problems but the is-
sues plaguing single-orbital problems remain. The LMA
is a highly14,15 accurate technique, that has been bench-
marked extensively16 with NRG, but the method has not
been used to study lattice problems except the periodic
Anderson model17 . Moreover, extensions to symmetry
broken phases or multiple orbital problems remain to be
carried out.
The IPT is a simple, second order perturbation the-
ory based method and it has been used widely to solve
impurity18,19 and lattice problems20 at zero as well as at
finite temperature. In the IPT, a self-energy ansatz is
constructed that interpolates between known limits (i.e.,
weak coupling, atomic and high frequency limits) which
is why it is also called an interpolative approach. It is
clear that even the single-orbital IPT is not free of ambi-
guities so different constraints or limits to construct the
ansatz yield different results. Hence, an IPT for multi-
orbital problems has been ‘synthesized’ in many different
ways by various groups21 , and we discuss these variations
next.
The IPT ansatz for the self-energy Σ(ω) is based on a
rational or continued fraction expansion of a specific sub-
set of diagrams, and consists of a small number of free
variables that are fixed by satisfying various limits, such
as atomic and high frequency limits and conservation
laws such as the Luttinger’s theorem. Such an interpola-
tive approach was first initiated by Martı́n Rodero18,22
for the single impurity and periodic Anderson models.
The approach used the second order self-energy as a
2
building block and the pseudo-chemical potential µ0 , was
fixed by assuming that the occupation n0 of the non-
interacting part of the Anderson impurity problem is
equal to the lattice occupation n. Soon after the devel-
opment of DMFT, the single band Hubbard model was
studied by Muller-Hartmann23 using self-consistent per-
turbation theory Σ[G]. The self-consistent perturbation
method is able to produce a coherence peak in the single
particle spectral function; however, it fails to reproduce
the high energy Hubbard bands. For the single impu-
rity Anderson model (SIAM), Yosida and Yamada19,24
demonstrated that perturbation theory in U is quite
well behaved for the symmetric case when expanding
around the Hartree-Fock solution. Based on these find-
ings, Georges and Kotliar20,25 introduced an impurity
solver called iterative perturbation theory (IPT) to solve
the single band Hubbard model within DMFT which is
based on mean-field G0 and is able to capture coher-
ent and incoherent features of single particle spectrum
quite successfully. This is one of the biggest advantages
of theories based on G0 rather than the fully dressed
G. Another advantage they offer is that, these theories
naturally avoid any form of two particle divergence and
are therefore able to provide a reliable description also
“beyond” the perturbative regime26 . This is very impor-
tant, because the twoparticle divergencies, which might
induce ambiguities in the numerical determination of the
DMFT self-energy (Σ[G]) within standard perturbation
theory schemes, occur in a rather large portion of the
phase diagram, including the metallic regime much be-
fore the Mott transition27 .
Subsequently, Kajueter and Kotliar28,29 proposed a
modification to the IPT called modified iterative per-
turbation theory (MIPT). In addition to the usual con-
straints of IPT, the MIPT constrains the zero frequency
behavior of the self energy by adding a pseudo chem-
ical potential µ0 to the Hartree corrected bath propa-
gators. This pseudo-chemical potential, µ0 , can be ob-
tained in different ways so there is an ambiguity in the
method. Kajueter28 fixed this free parameter by satisfy-
ing the Friedel’s sum rule (equivalently Luttinger theo-
rem), hence his method is called IPT-L. The Luttinger
theorem and Friedel’s sum rule are valid only at zero
temperature, hence for finite temperature calculations,
Kajueter21 used the same µ0 that was obtained at zero
temperature.
To study spontaneous magnetism in the single band
Hubbard model, Potthoff, Wegner and Nolting30,31 im-
proved MIPT further by taking into account the spectral
moments up to third order and instead of fixing µ0 by
using Luttinger theorem, they fixed it by the n = n0
constraint. This method may be called IPT-n0 . They
also considered the simpler option, where lattice chemi-
cal potential µ is equal to the pseudo chemical potential
µ0 . This is called IPT-µ0 and they bench-marked IPT-L
and IPT-n0 with IPT-µ0 . Recently, Arsenault, Sémon,
and Tremblay32 bench-marked IPT-n0 with CTQMC and
found the pathology in IPT-n0 that, in the strong cou-

pling regime, the method does not recover a Fermi-liquid
for filling close to n = 1. They suggested a new method
(IPT-D) to fix the µ0 through a double occupancy con-
straint. The range of schemes originating from the inher-
ent ambiguities at the single-orbital level give an idea of
the far larger range of approximations that can be built at
the multi-orbital level. These schemes will be described
next.
Kajueter21 extended his single orbital perturbative
scheme to the degenerate multi-orbital case. He used
the coherent potential approximation (CPA) to calculate
higher order correlation functions in the self energy. He
showed, by benchmarking against ED, that the scheme
provides reasonable results only if the total particle den-
sity per site is less than one. For fillings greater than one,
his scheme produced a false double peaked structure at
the Fermi level instead of a single resonance. The reason
for such a spurious structure is that the high frequency
tails in the continued fraction expansion can be system-
atically improved by considering poles involving higher-
order correlations functions in the self-energy, but this in
turn seriously degrades the low frequency behavior when
the Luttinger’s theorem is attempted to be satisfied. To
study quantum transport in mesoscopic systems such as
multi-level quantum dots, Yeyati et al.33 introduced an
interpolative scheme based on IPT-n0 . Liebsch34 ap-
plied an extension of IPT to study the orbital selec-
tive Mott-transition, using which he showed that inter-
orbital Coulomb interactions gives rise to a single first-
order transition rather than a sequence of orbital selec-
tive transitions. In Liebsch’s extension of IPT for the
multi-orbital case, he chose the self energy to be the com-
bination of Hartree term and second order pair-bubble
diagram with interaction vertices between electrons in
different orbitals on the impurity. Laad et al.35 con-
structed an interpolative scheme for multi-orbitals that
was used extensively to study real materials through the
LDA+DMFT framework. In a similar context, Fujiwara
et al.36 developed an interpolative approach for degener-
ate multi-orbitals. The novelty of their method was that
they used ligand field theory in the atomic limit to find
the higher-order correlation functions.
Although there exist a large range of schemes for ex-
tending IPT to the multi-orbital case, extensive bench-
marking of any single method has not been carried out.
Recently Savrasov et al.37 , and Oudovenko et al.38 de-
veloped an interpolative approach for degenerate multi-
orbitals based on a simple rational form of the self-energy,
where the unknown coefficients in the self-energy are de-
termined using slave boson mean-field and Hubbard I
approximations. In their Hirsch-Fye-QMC work on the
SU(4) Hubbard model, they have observed a good agree-
ment in the particle-hole asymmetric cases.
In the present work, we build upon the previous knowl-
edge to develop an interpolative scheme for solving a
general multi-orbital quantum impurity problem. Our
scheme is also based on the second-order self-energy
as a building block and we use the generic name for
3
the method as simply multi-orbital iterative perturba-
tion theory (MO-IPT). Our method has a single pseudo-
chemical potential µ0 , that is found by satisfying the Lut-
tinger’s theorem. We impose the correct high frequency
and atomic limits to get the unknown coefficients in the
self-energy ansatz. In the single orbital case, we find that
MO-IPT recovers the usual MIPT self energy expression
and for the degenerate multi-orbital case, our MO-IPT
self-energy expression reduces to that of Kajueter21 . The
main novelty lies in handling the high frequency poles in
a systematic way. The method is general enough that it
can be applied to study symmetry broken phases, Hund’s
coupling (density-density type) and crystal field effects.
Since MO-IPT is a semi analytical method it needs
to be bench-marked. Subsequent to the description of
the method, we embark upon an extensive benchmark-
ing of MO-IPT with numerically exact, hybridization ex-
pansion continuous time quantum Monte Carlo method
(S-CTQMC)39 as implemented in the ALPS40 libraries
and our own implementation of interaction expansion
CTQMC (W-CTQMC). Our main conclusion is that the
MO-IPT method works very well when used away from
integer-fillings, even at reasonably strong coupling. Us-
ing MO-IPT, we have addressed issues disputed in the
current literature of doped Mott insulators32 and cova-
lent band insulators41 . In the multi-orbital degenerate
case, the method proposed by Kajueter21 shows spuri-
ous features which restricted its applicability to fillings
smaller than one. However, our approach circumvents
all the above issues and moreover captures the filling de-
pendent effect of the Hund’s coupling42 in the low en-
ergy scale. Our study of crystal field effects (the non-
degenerate case) is the first attempt to extend the ansatz
beyond the degenerate case. In addition, we have shown
that our method produces a good default model for the
analytic continuation of CTQMC data using the max-
imum entropy method43 . We have also integrated the
MO-IPT with material-specific, density functional the-
ory based calculations, and thus tested it for a proto-
typical example of strongly correlated electronic system,
SrVO3 . A rather good agreement is obtained when the
MO-IPT photo-emission spectra (PES) is compared with
experiments.
We have organized the paper as follows. In section
II, we outline the formalism for MO-IPT. In section III,
we discuss results when the MO-IPT is applied in the
DMFT context for lattice problems. In section IV, we
present our conclusions along with future directions and
possible improvements.
II. Model and Formalism
The multi-orbital Hubbard model for density-density
type interactions and for cubic environment in standard
 4

second quantization notation is given by Ĥ(k))−1 is the hybridization matrix or equivalently the
X X αβ † XU self-consistently determined bath and Σ̂imp (ω + ) is the
H= iα niασ + Tij (ciασ cjβσ + h.c) + niασ niασ̄impurity self-energy obtained by solving the impurity
iασ iασ
2 problem. The set of equations is closed by noting
ijαβσ
X (U − 2J) X (U − 3J) that, within DMFT, the lattice self-energy is momentum-
+ niασ niβσ0 + niασ niβσ . independent and is the same as the impurity self-energy,
2 2 i.e Σ̂(k, ω + ) = Σ̂imp (ω + ). The local Green’s function ob-
iασ6=βσ 0 iσα6=β
(1) tained in Eq. (2) is used for defining a new hybridization
as
where c†iασ creates the electron at lattice site i, in orbital
ˆ + ) = (ω + + µ)I − ˆ − Σ̂imp (ω + ) − Ĝ−1 (ω + ) . (8)
∆(ω
α with spin σ and cjβσ annihilates the electron at site j, loc
in orbital β with spin σ. We are mainly interested in the
local single particle electron dynamics, which is given by
the momentum sum of the lattice Green’s function
Obtaining the self-energy however is the most challeng-
+
X 1 ing step, and we employ multi-orbital iterated perturba-
Ĝloc (ω ) = . (2)
(ω + + µ)I − Ĥ(k) − Σ̂(k, ω + ) tion theory to solve the multi-orbital Anderson model.
k
The starting point, as usual, is an ansatz for the impu-
+
Where ω = ω+iη and η is the convergence factor. Here rity self-energy, given by28
Ĥ(k) comprises intra-unit-cell hybridization and inter-  
unit-cell hopping, namely Σ̂ imp (ω) =
αβ
 
P (2)
Ĥ(k) = Ĥintra + Ĥ(k)inter (3) X Aα γ6=α Σαγ (ω)
  δαβ  Uαγ hnγ i + (2)
.
= iα δαβ + Tiiαβ
P
where Ĥintra (4) γ6=α 1 − Bα γ6=α Σαγ (ω)
αβ
  (9)
and Ĥ(k)inter = (k)αβ , (5)
αβ
The self-energy is thus restricted to being diagonal in
αβ the orbital basis. In the above ansatz, the first term is
where iα are orbital energies, Tii are intra-unit cell hy-
simply the Hartree energy and the second term contains
bridization matrix elements, and (k)αβ is the dispersion
the second order pair-bubble diagram Σ̂(2) of matrix size
of the lattice, that depends on its geometry. For exam-
N×N, where N is the number of orbitals. The second
ple, in the case of a simple cubic lattice, (k)αβ assumes
order pair-bubble diagram on the real frequency axis is
the form, −2tαβ ij (cos kx + cos ky + cos kz ). given by
Within DMFT, one can map the multi-orbital Hub- Z Z Z
bard model on to an auxiliary impurity problem with (2)
Σαβ (ω) =Uαβ 2
d1 d2 d3 ρα (1 )ρβ (2 )ρβ (3 )
a self consistently determined bath. The Hamiltonian of
the corresponding single impurity multi-orbital Anderson nF (−1 )nF (2 )nF (−3 ) + nF (1 )nF (−2 )nF (3 )
model, is expressed in standard notation as: ,
ω + − 1 + 2 − 3
1 X   (10)
Vkα c†kα fα + fα† ckα
X
Himp = (α − µ)fα† fα +
N
α k,α where ρα = − π1 ImG̃αα and G̃αα is the Hartree corrected
1 bath propagator, which is obtained from a Dyson like
kαβ c†kα ckβ +
X X
+ Uαβ nα nβ (6) equation, and is given by
2
k,α,β α6=β
 
Here α and β are impurity orbital indices including spin. G̃ −1 = Ĝ−1 ˆ − (µ − µ0 ) I .
loc + Σ̂ +  (11)
The first term in the above equation represents the or-
bital energy; the second term is the hybridization be- The pseudo chemical potential, µ0 , is found at T = 0 by
tween the impurity and the host conduction electrons, satisfying the Luttinger’s theorem,
the third term represents the host kinetic energy and !
Im 0
Z
the final term is the local Coulomb repulsion between dΣ̂(ω)
− dωTr Ĝimp (ω) = 0 . (12)
electrons at the impurity. The corresponding impurity π −∞ dω
Green’s function is given by,
At finite temperature, an ambiguity exists in the deter-
1 mination of the pseudo-chemical potential. We choose to
Ĝimp = , (7)
+ ˆ
(ω + µ)I − ˆ − ∆(ω + ) − Σ̂imp (ω + ) use the µ0 determined at zero temperature for all finite
temperatures. The chemical potential, µ, is found by fix-
where (ˆ ˆ + ) = P |Vkα |2 (ω + I −
)αβ = α δαβ . Here ∆(ω ing the total occupancy from the local Green’s function,
k

Ĝloc , to be equal to the desired filling,
Z 0
1 AG (ω)dω
− Im TrĜloc = ntot , (13)
π −∞
where the trace is over the spin and orbital indices. The
unknown coefficients Aα , Bα from Eq. (9) are obtained in
the standard way by satisfying the high frequency limit
and the atomic limit respectively. The detailed procedure
to derive Aα , Bα and their expressions are discussed in
the Appendix. These coefficients contain higher order
correlation functions. The order of the correlation func-
tions depends on number of poles in the self energy. For
example a pole of order n involves (n+1)th order correla-
tion functions. For a two pole ansatz Aα and Bα involve
two and three particle correlation functions. We calculate
the two particle correlation function44 using the equation
of motion method to obtain45
Z
X 1
Umm0 hnm nm0 i = − dωnF (ω)Im [Σm (ω)Gm (ω)] .
π
m0 6=m
(14)
This single equation is not sufficient to find all the two-
particle correlators. Hence as an approximation, we use
the following:
R We study the above model within DMFT for a semi-
dωnF (ω)Im(Σm (ω)Gm (ω))
hnm nm0 i = − . (15)
πUmm0 (Norb − 1)
We calculate the three particle correlation function en-
countered in Bα approximately by decoupling it in terms
of two and single particle correlation functions. In this
work, we have ignored the three particle correlation func-
tion.
III. Results and Discussion
The formalism developed in the previous section is ap-
plied to a wide variety of correlated systems. We begin
with a discussion of the well studied paramagnetic Mott
transition in the half-filled single-band Hubbard model.
Then we examine the doped Mott insulator. The covalent
insulator is considered next, followed by the two-orbital
Hubbard model. For the latter, We investigate the effect
of filling, Hund’s coupling and crystal field splitting. Fi-
nally, we move to real material calculations considering
specifically the case of SrVO3 . As mentioned earlier we
bench-mark our results with those from numerically ex-
act CTQMC39 methods. The CTQMC formalism yields
results on the Matsubara frequency axis so to get the
real frequency data, analytical continuation is required.
We avoid analytic continuation by transforming the real
frequency data obtained from MO-IPT to imaginary fre-
5
quencies using the following spectral representations:
Z
G(iωn ) = , (16)
iωn − ω
and
Z
AΣ (ω)dω
Σ(iωn ) = , (17)
iωn − ω
where AG (ω) = −ImG(ω)/π and AΣ (ω) = −ImΣ(ω)/π.
In order to quantify the efficiency of the method, the
imaginary part of the self energy needs to be bench-
marked rather than the Green’s function. This is because
the former is far more sensitive than the latter and more-
over, the low energy scale of the system depends on the
imaginary part of the self energy.
A. Single band Hubbard model: Half-filled case
The Hamiltonian for the single band Hubbard model
is given by
XU
Tijσ (c†iσ cjσ + h.c) +
X X
H= iσ niσ + niσ niσ̄ .
ijσ iσ iσ
2
(18)
elliptical density of states, given by
s 
2
4 W
ρ() = − 2 . (19)
πW 2 2
Here W is the full-band width. In our calculations, we
choose the energy unit to be W 2 = 1.
The half-filled Hubbard model exhibits an interaction-
driven metal-insulator Mott transition at a critical Uc .
Terletska et al.46 found that the critical exponents and
scaling functions obtained by IPT are identical to those
from CTQMC. Here, we revisit this case and benchmark
the quasiparticle weight, double occupancy, spectra and
imaginary part of the self-energy. The MO-IPT method
reduces to the second order perturbation theory in terms
of Hartree-corrected propagators. In Figure 1(a) we com-
pare the quasi-particle weight Z obtained from different
impurity solvers and several values of the Coulomb inter-
action. The values of Z obtained from S-CTQMC match
well with those from NRG47 for all values of U/W except
close to the Mott-transition. This is most likely because
we have done CTQMC calculations at β = 64, while NRG
is at zero temperature. The critical interaction strength,
Uc 48
W ≈ 1.35 obtained from both the methods agrees very
well. The Z obtained from MO-IPT at β = 64 matches
quantitatively with CTQMC and NRG in the weak cou-
pling limit and only qualitatively in the proximity of the
transition. On the other hand, the results of the self en-
ergy functional approach (SFA)49 agree with MO-IPT in

FIG. 1. (Color online) (a) Quasi-particle weight Z of the
single band half-filled Hubbard model obtained with differ-
ent impurity solvers (see text for more details) (b) Double
occupancy D obtained from MO-IPT and S-CTQMC.
the strong coupling limit rather than in the weak cou-
pling limit. The MO-IPT yields the critical value of
Uc
W = 1.42, which is in good agreement with the critical
value UW = 1.45 obtained from SFA
c 49
at zero tempera-
ture. The double occupancy obtained from MO-IPT and
S-CTQMC (shown in panel (b) of Fig. 1) also match,
except very close to the transition. A detailed compar-
ison of spectra from S-CTQMC and W-CTQMC with
the same from MO-IPT (transformed to imaginary fre-
quencies) is shown in Figure 2. The left panels show the
imaginary part of the Green’s function at U/W = 1.0
(top panel) and U/W = 1.5 (bottom panel), while the
right panels show the imaginary part of the correspond-
ing self-energies. The excellent agreement between the
three methods is clearly evident.
B. Single band Hubbard model: Doped Mott
insulator case
The single band Hubbard model has gained a lot
of interest, because the doped Mott insulator regime
is believed to capture the essential physics of high Tc
superconductors51 . This regime is, in reality, highly
complex, because many different factors such as prox-
imity to the antiferromagnetic Mott insulator, disor-
der, d-wave superconducting fluctuations and pseudogap
physics have to be treated on an equal footing. Hence,
investigations of the doped Mott insulator in all its glory
6
FIG. 2. (color online) Comparison of the imaginary part
of Matsubara Green’s function (left panels) and self energy
(right panels) obtained from MO-IPT, S-CTQMC and W-
CTQMC50 for U/W = 1.0 (top panels) and U/W = 1.5 (bot-
tom panels) at β = 64.
represents one of the toughest challenges in condensed
matter. Here, we take a simplistic approach to the prob-
lem, and investigate the performance of MO-IPT in the
paramagnetic doped Mott insulator in infinite dimen-
sions. Our MO-IPT reduces basically to the IPT-L in
this regime.
A comparison of quasi-particle weight at U/W = 1.5
obtained from MO-IPT and S-CTQMC as a function of
filling (Fig. 3) yields, surprisingly, an excellent agree-
ment. We observe that as we decrease the filling (from
1) for a given U/W , the Mott insulator turns into a
strongly correlated metal and finally ends up as a sim-
ple metal. In the strong coupling limit, for filling close
to n = 1, the IPT-n0 method gives an insulating solu-
tion, while the IPT-L correctly predicts a metal in agree-
ment with exact methods. Kajueter and Kotliar21 have
benchmarked the real-frequency spectral functions ob-
tained from IPT-L with exact diagonalization calcula-
tions and had found good agreement. We find that the
imaginary part of the Green’s function and self-energy
obtained from IPT, when transformed to the Matsubara
frequency axis using Eqs. (16), (17) are almost identical
to those obtained from the strong coupling and weak-
coupling variants of CTQMC (see Fig. 4). The slope of
the ImΣ(iωn ) as ωn → 0 is 1 − 1/Z, and the good agree-
ment of Z shown in Fig. 3 is simply a reflection of the
detailed agreement for all frequencies. Such an excellent
agreement is truly surprising because IPT is a perturba-
tive method by construction and the strongly correlated,

FIG. 3. (color online) Quasi-particle weight obtained from
MO-IPT (or IPT-L) is compared to the same obtained from
CTQMC for the paramagnetic doped Mott-insulator as a
function of filling with U/W = 1.5 and β = 64.
FIG. 4. (color online) Doped Mott insulator: Comparison of
imaginary part of Matsubara Green’s function and self energy
obtained from MO-IPT, W-CTQMC and S-CTQMC for U/W
= 1.5 at different fillings and β = 64.
7
doped Mott insulator regime should not, in general, be
amenable to perturbative methods.
C. Covalent Insulator:
The discovery of topological insulators52 has led to a
renewed interest in the role of e-e correlations in band
insulators (BI)53 . The prime examples of such materials
would be FeSi54 and FeSb2 55 , since experimental mea-
surements indicate a small optical gap and large ther-
mopower (at low T ). Increasing temperature leads to
closing of the gap, and concomitantly a insulator-metal
crossover in the resistivity. Such large scale spectral
weight transfers are highly indicative of strong corre-
lations. Specific heat measurements also seem to vali-
date this observation. The band gap in these systems
is a simple consequence of the structure of the hopping
matrix and not of completely filled electronic shells41 .
Hence these materials are called covalent insulators41,53 .
A Hamiltonian that describes the covalent insulator is
given by41
  X
X † a

H= ak,σ b†k,σ Ĥ(k) k,σ + Uαα niα niα ,
bk,σ
kσ iα
(20)
where α = a and b are two sub-lattices with semi-elliptic
bands and having dispersion k and -k respectively. The
two sub-lattices are coupled by a k−independent hy-
bridization V . While the unit of energy is chosen to be W
= 2 throughout, for this subsection W=4 has been chosen
in order to benchmark with earlier results41 . This is the
first two-band model we have studied in this work, since
the previous cases pertained to the single-band Hubbard
model. Hence this will be the first real test of the ‘multi-
orbital’ part of MO-IPT. Since this is still the half-filled
case, Luttinger’s theorem does not have to be satisfied
explicitly. The Aα = 1 and Bα = 0 for all orbitals.
Thus, the MO-IPT used for the covalent insulator case
is equivalent to that employed by Liebsch34 for studying
the Mott transition in the two-band Hubbard model.
Before presenting the numerical results, we will present
a simple analytical argument about the absence of any
intermittent metallic phase in the CBI at zero tempera-
ture unlike the interaction-induced metallic phase in the
Ionic Hubbard model56 . The structure of Hσ (k) in the
CBI is given by,
 
k V
Hσ (K) = , (21)
V −k
and the corresponding impurity Greens function for sub-
lattice ’a’ is given by,
ζbσ (ω + , )ρ0 ()
Z
+
Gaσ (ω ) = d , (22)
ζaσ (ω + , )ζbσ (ω + , ) − V 2
 8

where

ζaσ (ω + , ) = ω + iη + µ −  − Σaσ (ω + ) ,
ζbσ (ω + , ) = ω + iη + µ +  − Σbσ (ω + ) ,

and η → 0+ . In the half-filling case, the Hamiltonian has

mirror type symmetry between sublattices, which reflects
in the impurity Green’s function and self-energy in the
following way,
∗
Gaσ (ω + ) = − Gbσ (−ω + ) ,

(23)
+ + ∗
 
Σaσ (ω ) = U − Σbσ (−ω ) . (24)

By using above self-energy symmetry relation, it is easy

to see that

ζaσ (ω + , ) = −[ζbσ (−ω + , )]∗ , (25)

then Eq. (22) can be written as,

∗
(−ω + , )ρ0 ()
Z
+ ζaσ
Gaσ (ω ) = d ∗ (−ω + , ) + V 2
. (26)
ζaσ (ω + , )ζaσ

With the assumption that in the band insulator and
metallic phases, a Fermi-liquid expansion of self-energy
ω→0
holds, namely that Σ(ω) → Σ(0) + ω(1 − 1/Z) + O(ω 2 ).
Then, the value of imaginary part of self-energy at zero
frequency (and zero temperature) is ImΣaσ (0) = 0, and
the corresponding density of states (DOS) Daσ (0) =
− π1 ImGaσ (0) is given by,
dρ0 () πη
Z
Daσ (0) = , (27)
η 2 + [Re(ζaσ (0, ))]2 + V 2
where Re(ζaσ (0, ))=[µ −  − ReΣaσ (0)]. In the limit η →
0+ ,
Z p  same obtained through strong coupling CTQMC (on the
Daσ (0) = dρ0 ()δ V 2 + (µ − ReΣaσ (0) − )2 .
Since the argument of the Dirac delta function is posi-
tive definite for any V 6= 0, the density of states at the
chemical potential, Daσ (0) will necessarily vanish for any
V 6= 0, and hence the system will be gapped. Thus for
the CBI, interactions do not close the gap, no matter how
strong they are. This implies a clear absence of metallic-
ity in these insulators.
FIG. 5. (color online) Covalent insulator: (a) Quasi-particle
weight Z as a function of U/W obtained from MO-IPT (black
circles) and CTQMC (red squares) for β = 60 and V=0.5.(b)
Double occupancy as a function of U/W obtained from MO-
IPT and S-CTQMC. (c) Charge gap as a function of U/W
obtained from MO-IPT at T=0.
gap with decreasing temperature. Precisely this behavior
is seen in the real frequency spectra (left panels, Fig. 6),
which arises from the spectral weight transfer in the self-
energy as a function of temperature. The high reliabil-
ity of these spectra and self-energies computed through
MO-IPT is apparent in the excellent agreement with the
Matsubara axis, in Fig. 7). The crossover of the band-
insulator to Mott insulator is also visible in the increas-
ing (negative) slope of the imaginary part of self-energy
with increasing U/W . Similar kind of correlation induced
transitions have also observed between topologically triv-
ial and non-trivial band insulators58,59 .
D. Two orbital Hubbard model:

The quasi-particle weights (Fig. 5(a)) and double occu-
pancy (Fig. 5(b)) of sublattice ’a’ obtained from MO-IPT
and S-CTQMC (shown as black circles and red squares
respectively) are in close agreement except in the prox-
imity of the transition of the correlated band insulator
to a Mott insulator. Unlike the ionic Hubbard model
case57 , we do not see any intervening metallic phase be-
tween the correlated band insulator and the Mott insu-
lator. This is also consistent with the S-CTQMC and
analytical results. At high temperatures, the correlated
band insulator should be gapless, and must develop the
Encouraged by the excellent benchmarking of MO-
IPT with CTQMC for the two-band covalent insula-
tor system, we now move on to the two-orbital Hub-
bard model60,61 . The Hamiltonian, in standard nota-
tion, for a cubic environment and for unbroken spin sym-
metry, is described in Eq. (1). Throughout the paper,
we have considered local interactions of density-density
type which are obtained by neglecting spin flip and pair-
hopping terms that must be present for a rotationally
invariant Hund’s coupling. The hopping is taken to be
diagonal in orbital indices for simplicity.

FIG. 6. Covalent insulator: Spectral functions (left panels)
and imaginary part of self energy of sublattice a (right panels)
from MO-IPT at U/W = 1.25 and V=0.5 for a range of β =
1/T values (increasing T from top to bottom).
FIG. 7. (color online) Covalent insulator: Comparison of the
imaginary part of Matsubara (a) Green’s function and (b)
self-energy of sublattice ’a’ obtained from MO-IPT (black)
and S-CTQMC (red) for various U/W values and β=60.
9
(a) Half-filling: J = 0
We begin by considering the half-filled case (total occu-
pancy is two) with J = 0. The Hamiltonian (Eq. (1)) has
SU(4) symmetry in this situation. We have employed a
semi-elliptic non-interacting density of states of full-band
width W = 2 for the MO-IPT-DMFT calculations.
FIG. 8. (color online) (a) Two-orbital SU(4) symmetric Hub-
bard model at half-filling: Quasi particle weight as a function
of U/W obtained from different impurity solvers(SFA,MO-
IPT,FLEX,S-CTQMC at β = 64 and ED at T=0.0).(b) Dou-
ble occupancy obtained from MO-IPT (black circles) and hy-
bridization expansion CTQMC (red squares) for β=64.
In Fig. 8a, we plot the quasi-particle weight (Z) ob-
tained from different impurity solvers for the particle-
hole symmetric case. The results from strong cou-
pling CTQMC, ED62 and SFA49 , including the critical
Uc , where the system transitions from metal to Mott-
insulator, are in good agreement. The critical value Uc
obtained in the multi-orbital case is greater than the
value obtained in the single band case. The Mott tran-
sition is absent in the FLEX result13 . The MO-IPT is
seen to underestimate the quasiparticle weight as com-
pared to the other methods (except MO-IPT<nn>=0;
see below). However, the critical Uc agrees reasonably
well with that from hybridization expansion CTQMC.
The green diamonds are from a variant of MO-IPT (used
e.g. by Fujiwara et al.36 ) where the two-particle correla-
tion function is simply decoupled into two single-particle
terms (hnα nβ i=hnα ihnβ i). The neglect of two particle
correlations leads to a much worse comparison than MO-
IPT. In contrast to the not-so-good agreement with exact
methods for the quasiparticle weight, the average double

occupancy obtained from MO-IPT shows excellent agree-
ment with CTQMC (see Fig. 8b). Since the total energy
of the system depends on single particle and two parti-
cle correlation functions, we expect that thermodynamic
quantities like total energy or specific heat computed
through MO-IPT might be reliable. One more impor-
tant observation is that the double occupancy remains
finite and almost constant even beyond the Mott transi-
tion, unlike the the single band case. We also compare
the single-particle Green’s function and self-energy on
the Matsubara frequency axis (Fig. 9). At high frequen-
cies, the agreement between MO-IPT and S-CTQMC is
seen to be excellent, while the agreement worsens at low
frequencies, especially with increasing U/W .
FIG. 9. (color online) Two-orbital SU(4) symmetric Hubbard
model at half-filling: Imaginary part of Matsubara Green’s
function (left panels) and self energy (right panels) obtained
from MO-IPT (red solid lines) and S-CTQMC (black solid
lines) at β=64.
(b) Half-filling: Effect of Hund’s coupling (J)
The interplay of Hund’s coupling, J, and local inter-
action, U , has been investigated by several groups. The
main consensus is that strong correlation effects can be
affected significantly through J 42,62 . For example, in the
half-filled case, the Uc for Mott transition is lowered by
(N − 1)J, where N is the number of orbitals, while the
critical U is enhanced by 3J in the non-half-filled (but in-
tegral occupancy) case62 . It is important to know the ex-
tent to which the interplay between J and U is captured
by the MO-IPT method. In Fig. 10(a) and (b), the quasi-
10
FIG. 10. (color online) Two orbital half-filled Hubbard model,
finite J. Quasi particle weight dependence on U/W obtained
from (a) strong coupling CTQMC, (b) ED and (c) MO-IPT
for various J values. Insets in the panels (a) and (b) show the
effect of J on Z in the weak coupling regime.
particle weight Z for different values of J obtained from
S-CTQMC and ED62 is shown. Indeed, with increasing
J, the Uc at which Z → 0 decreases sharply, as expected
from the atomic limit. Also, for each J, the quasiparticle
weight decreases monotonically with increasing interac-
tion strength. Although the latter trend is qualitatively
captured by the MO-IPT result (shown in Fig. 10c) for
larger J, there is a disagreement with the exact results at
lower J values. The MO-IPT yields a Uc that is a non-
monotonic function of J. The insets of panels a and b in
Fig. 10 zoom in on the low interaction (U/W . 0.3) part
of the main panels. Unlike for U/W & 0.3, where increas-
ing J leads to a monotonic reduction of Z, a rise and fall
of Z is observed for U/W . 0.3. Although such a trend is
achieved by MO-IPT as well, the non-monotonicity sus-
tains even for larger U/W . A frozen local-moment phase
is seen in the S-CTQMC calculations for any given J in
the strong coupling limit, while such a phase is not ob-
served either by ED62 or MO-IPT calculations. It must
be mentioned here that the CTQMC calculations employ
a density-density type Hund’s coupling, while the ED
employs a fully rotationally invariant J. Although the
quasiparticle weight dependence on U and J is not accu-
rately captured by MO-IPT, the single-particle dynamics
on all scales is in qualitative agreement with S-CTQMC
calculations (as seen in Fig. 11).
 11

FIG. 11. (color online) Two orbital half-filled Hubbard model,
finite J. Imaginary part of Matsubara Green’s functions
(left panels) and self-energy (right panels) obtained from S-
CTQMC (black) and MO-IPT (red) for different values of J
and U/W at β=64.
(c) Away from Half-filling: Effect of J
FIG. 12. (color online) Two-orbital Hubbard model: Effect
of J away from half-filling (ntot = 1.1). The imaginary part
of the Matsubara self-energy for various J-values, and fixed
U/W = 1 as computed within (a) S-CTQMC and (b) MO-
IPT. Comparison of quasi particle weight obtained from MO-
IPT (black circles) and CTQMC (red squares) as a function
of U/W for (c) J = 0.0, (d) J = U/4 and (e) J = U/3.5 for
β = 64; and (f) as a function of J for a fixed U/W = 1.0.

The MO-IPT method works best away from half-filling,

which is consistent with the results of comparisons car-
ried out previously by other groups63 . In order to illus-
trate this, here we study the two orbital Hubbard model
for a total occupation of ntot = 1.1. The imaginary part
of the Matsubara self-energy obtained from S-CTQMC
matches well with that from MO-IPT (Fig. 12, panels
(a) and (b)), hence the latter does well in this regime.
This observation is reinforced by the panels (c)-(e), which
show a comparison of the quasiparticle weights as a func-
tion of U/W for three values of J, namely J = 0, U/4.0
and U/3.5. The results of MO-IPT are seen to agree
very well with those from CTQMC. For most real ma-
terial calculations, the regime considered in this sub-
section is perhaps the most relevant. Hence, accurate
results from MO-IPT prove its efficacy for integration
into first-principles approaches. The Hund’s coupling
and Coulomb interaction have a synergistic effect at half-
filling, while in the doped case, the reverse occurs42 . This
is shown in panel (f) of Fig. 12, where an increase of Z
is seen with increasing Hund’s coupling at a fixed inter-
action strength. It is quite instructive to study the real
frequency spectral functions and self-energies as obtained FIG. 13. (color online) Two-orbital Hubbard model away
from MO-IPT. These are shown in Fig. 13 for various val- from half-filling. Real frequency spectral functions (left pan-
els) and minus imaginary part of self energy (right panels) for
ues of interaction strength and Hund’s coupling, J. In
various U/W and J values.
the absence of Hund’s coupling, the spectrum (shown in

the left panels of Fig. 13) exhibits spectral weight trans-
fers characteristic of increasing correlation strength: a
central resonance that becomes sharper, and Hubbard
bands that grow in prominence with increasing U/W .
However, at a fixed U/W , increasing Hund’s coupling
leads to a reversal of the aforementioned trend, i.e, a
broadening of the resonance and a melting of the Hub-
bard band (see e.g. left panel bottom figure). In this
parameter regime, a previous formulation of the multi-
orbital iterated perturbation theory63 found a double
peak structure at the chemical potential. Such a feature
was shown by the authors28 to be spurious by comparison
to results from exact diagonalization. The reason we do
not observe such a spurious feature is that we have con-
sidered only two poles in the self-energy, in contrast to
the formulation of Ref. 28, where they have retained all
the eight poles (for a two-orbital model). Although our
ansatz seems like an ad-hoc truncation scheme, the justi-
fication for such a scheme lies in its excellent agreement
with CTQMC results (shown in Fig. 14) and the absence
of spurious features. In Fig. 14, the imaginary part of
Matsubara Green’s functions and self energies obtained
from MO-IPT are compared with those from CTQMC
for three values of J at U/W = 1.25 and β =64. For all
values of the Hund’s coupling, an excellent agreement is
obtained.
FIG. 14. (color online) Two-orbital degenerate Hubbard
model away from half-filling (ntot = 1.1). Comparison of the
imaginary part of the Matsubara Green’s function (left pan-
els) and the self energy (right panels) obtained from MO-IPT
and S-CTQMC for various values of J at U/W = 1.25.
12
E. Two orbital Hubbard model: Crystal field
splitting and Hund’s coupling
We now proceed to the case of a non-degenerate two-
orbital model with crystal field splitting64 in the pres-
ence of Hund’s coupling. In most materials, the crys-
talline environment lifts the orbital degeneracy65 . For
example in transition metal oxides, due to crystal field
effects, the five fold degenerate d-level splits into triply
degenerate t2g and doubly degenerate eg levels and the
corresponding energy gap is ∼1-2 eV. The degeneracy
of each of these levels (t2g , eg ) is further lifted by dis-
tortions such as the ones occurring in GdFeO3 , or aris-
ing through the Jahn-Teller effect or spin-orbit coupling.
The energy cost for such distortion induced splitting is a
few meV. Recently, Pavarini et al.66 studied crystal field
effects in d1 type perovskites such as SrVO3 , CaVO3 ,
LaTiO3 and YTiO3 . It was found that crystal field ef-
fects and cation-covalency (GdFeO3 -type distortion) lift
the orbital degeneracy and reduce the orbital fluctua-
tions. Thus, investigating crystal field effects in model
Hamiltonians67,68 is highly relevant for understanding of
real materials.
We have investigated the Hamiltonian in Eq. (1) by
considering two orbitals with energies 1 = 0.0 and
2 = −0.2W , which corresponds to a crystal field split-
ting of 0.2W . The results from MO-IPT, for a fixed to-
tal filling of ntot = 1.1, are compared with those from
strong coupling CTQMC at the corresponding orbital
occupancies. In Fig. 15, we compare the quasi particle
weights of the two orbitals obtained from MO-IPT with
that of CTQMC. We observe a better agreement of Z
for orbital-1 than for orbital-2. This must be expected,
since orbital-1 is further away from particle-hole symme-
try than orbital-2. The corresponding orbital occupan-
cies as a function of increasing interaction (and hence
J) are shown in Fig. 15. The deviation between results
from the two methods increases with increasing U and
J(= U/4), which indicates that MO-IPT is almost exact
for U/W . 0.5.
Next, we benchmark the single-particle dynamics in
the presence of crystal field splitting. In Fig. 16, we show
the imaginary part of the Matsubara frequency self ener-
gies obtained from MO-IPT and CTQMC for orbitals-1
and 2 (left and right panels respectively). The agreement
between the results is quite evident, this suggesting that
the MO-IPT should serve as a good method to study in-
teracting, real material systems with finite crystal field
effects and Hund’s couplings. This is especially true if
the material in question has a large number of bands,
which would make it prohibitively expensive to treat with
CTQMC, while MO-IPT would be able to handle it with
ease. We now demonstrate the efficacy of MO-IPT when
applied to a well studied, real material system, namely
SrVO3 .

FIG. 15. (color online) Crystal field effects. Quasi particle
weights for (a) orbital-1 and (b) orbital-2, obtained from MO-
IPT and CTQMC for various U/W values with J = U/4 at
β=64. The insets show the corresponding occupancies.
FIG. 16. (color online) Crystal field effects. Comparison of
the imaginary part of the self energy for orbital-1 (left) and
orbital-2 (right) obtained from MO-IPT and S-CTQMC for
various values of U/W and J = U/3.5.
13
F. Application to real materials: SrVO3
Over the past decade or so, the combination of density
functional theory (DFT) with dynamical mean field the-
ory, such as LDA+DMFT69 , has emerged as one of the
most powerful methods for electronic structure calcula-
tions of strongly correlated electronic systems. Although
the DFT results contain rich, material specific informa-
tion, being a single particle theory, it works well only for
weakly correlated systems where the ratio of Coulomb in-
teraction (U ) to bandwidth (W ) is small i.e., U/W  1.
If we consider the opposite limit of U/W  1, we have
successful methods like the Hubbard-I and Hubbard-III
approximations or the LDA+U method for predicting the
ground state of the system. But these also have limita-
tions, such as the neglect of dynamical fluctuations in the
LDA+U method. In nature, there are many materials,
for example transition metal oxides, which lie in between
these two limits. It has been established in the context
of model Hamiltonians that the DMFT can handle both
limits quite efficiently. Hence a natural combination of
LDA with DMFT is expected to bring predictive capa-
bilities in the theory of strongly correlated electronic sys-
tems. Nevertheless, LDA+DMFT is not without its own
bottlenecks.
One of the central issues of the LDA+DMFT method
is the correct definition of a correlated subspace. The
basic idea of a correlated subspace is to make an appro-
priate choice of energy window around the Fermi level
and fit the band structure to a few-orbital tight-binding
model. Many techniques have been proposed to con-
struct such a material specific ‘non-interacting’ Hamil-
tonian. The two major techniques for this purpose are
down-folding70 and projection based Wannier function
technique71 . In general, bands which are crossing the
Fermi level are considered in the desired energy window
for Hamiltonian construction. For example in transition
metal compounds bands having d-orbital character nor-
mally cross the Fermi level. This process becomes simple
if there is no hybridization in the system and these bands
with d-orbital character are well separated from other
bands like the p-bands. As Dang et al.72 pointed out, a
mixing of these d orbital bands with p orbital bands can
create several complications.
After getting the ’non-interacting’ Hamiltonian , one
can add various types of interactions terms to obtain a
full material-specific multi-orbital model. The solution
of such a Hamiltonian is however a major challenge and
this is where the MO-IPT can be most useful, since it
scales only algebraically with increasing number of bands,
while yielding real frequency quantities directly. In con-
trast, impurity solvers like CTQMC and ED scale expo-
nentially with increasing number of orbitals and are very
expensive, especially for investigations of real materials.
As a test case, we study SrVO3 which is considered a
prototypical example of a strongly correlated electronic
system.
 14

(a). Computational Details

We perform our density functional theory (DFT) cal-

culations with linearized augmented plane wave (LAPW)
based method as implemented in the all-electron package
WIEN2K73 . The experimentally determined structure74
of cubic SrVO3 in a non-magnetic phase was used for
the calculations (neglecting spin-orbit coupling). The
product of the plane-wave cut off (Kmax ) and the small-
est atomic sphere radius (RM T ) was chosen as RM T ×
Kmax = 7.0 for controlling the basis set. The radii of
the muffin-tin spheres were chosen to be 10 − 15% larger
than the corresponding atomic radii. Thus, the values
used for RM T were 2.50 for Sr, 1.89 for V and 1.71 for O.
With these parameters, charge leakage was absent and
our DFT results agree well with DFT calculation using
other basis sets75 . We utilize the generalized gradient
approximation (GGA) of Perdew, Burke and Ernzerhof76
for the exchange and correlation functional. In this calcu-
lation, we consider 512 k-points in the irreducible part of
the Brillouin zone. After getting the Bloch-eigen states,
all the necessary inputs for constructing the maximally
localized Wannier functions (MLWFs) are prepared by FIG. 17. (color online) Band structure of SrVO3 obtained
the WIEN2WANNIER code77 . Finally, the Hamiltonian from DFT.
HDFT is constructed in the maximally localized Wannier
basis by taking a projection of the three V − t2g orbitals
within the energy window of -1.0 eV to 1.8 eV with the
standard procedure implemented in Wannier9078 . We
begin by discussing the DFT results.

(b). GGA+DMFT: Results and discussion

Our computed band structure and density of states

(DOS) are presented in Fig. 17 and Fig. 18. The three
bands, crossing the Fermi level, are highlighted in cyan,
violet and grey colors. These bands originate from the
V −t2g states, and are located between -1.1eV and 1.5eV.
The V −eg states lie at higher energies, between 1.1eV to
5.8eV (see the projected density of states in Fig. 18). The
band structure agrees well with previous results by Ishida
et. al.79 obtained in the LAPW basis. When compared
with results from the linear muffin-tin orbital (LMTO)
calculations of Nekrasov et al.80 , the position of V − t2g
bands agrees well but the position of V − eg states differs
by about 0.3 eV. This discrepancy is, most likely, due to
the difference in basis sets used in the two calculations.
FIG. 18. (color online) The projected density of states (DOS)
A significant computational simplification results from
of SrVO3 as calculated by GGA (LAPW).
ignoring the hybridization between V − t2g and V − eg
orbitals, since the low energy correlated subspace com-
prises just the three V − t2g orbitals. Thus, the DFT
results yield a ‘non-interacting’ Hamiltonian ĤDF T (k), where Ĥint is the interaction term is given by
which in this case is a 3 × 3 matrix for each k. Thus, the X X
full DF T + DM F T Hamiltonian is given by Ĥint = U niα↑ niα↓ + (U 0 − δσσ0 J)niασ niβσ0 .
i,α iα6=β,σσ 0
Ĥ = ĤDF T (k) + Ĥint , (28) (29)

FIG. 19. (color online) Comparison of spectral function of
SrVO3 obtained from different methods for U = 3.44 eV and
J = 0.46 eV (see text for details).
In the above expression, i stands for V sites and α is
the t2g orbital index with spin σ. U , U 0 (= U − 2J)
and U 0 − J(= U − 3J) are the local, intra orbital and
inter orbital Coulomb repulsion respectively and J is
the Hund’s exchange. The local, non-interacting lattice
Green’s function, in the orbital basis, (Ĝ0 (ω)), can be
obtained from the DFT calculated ĤDF T (k) by the fol-
lowing equation as
Ĝ0 (ω) =
X h i−1 
+ chosen pseudo-chemical potential, the Luttinger’s inte-
(ω + µ)I − ĤDF T (k) − ĤDC (30)
k gral, Eq. (12) is computed using the converged self-energy
h i−1
ˆ
≡ (ω + + µ)I − ∆(ω) , (31)
where µ is the chemical potential and ∆(ω) ˆ is the hy-
bridization. In the DFT approach electronic correlations
are partially entered through the LDA/GGA exchange-
correlation potential. This part of the interaction (Ĥ DC )
has to be subtracted in the LDA+DMFT approach to
avoid double-counting. This is not an important issue
when the low energy effective Hamiltonian contains only
the d-manifold because we can absorb it into the chemi-
cal potential. However it is an important issue when the
low energy effective Hamiltonian contains O-2p orbitals
also. Various schemes for finding the double-counting
correction Ĥ DC exist, each with a different physical mo-
tivation. Details about such schemes may be found in the
work by Lechermann et al.81 and Nicolaus Parragh82,83 .
15
In general we can construct the modified host Green’s
function for the αth orbital as
h i−1 
G̃α = Ĝ−1 0 + 
ˆ + Ĥ DC − (µ − µ0 ) I . (32)
αα
We find the pseudo-chemical potential using the same
procedure as in the model calculations. The self-energy
can be found, e.g. through the MO-IPT method out-
(2)
lined in Section II. The second-order self-energy Σαβ in
Eq. (9)nis aofunctional of the modified host Green’s func-
tions, G̃α . The full local Green’s function for the lat-
tice Hamiltonian (Eq. (28)) is given by
Xh i−1
Ĝ = (ω + + µ)I − ĤDF T (k) − ĤDC − Σ̂(ω) .
k
(33)
The above Green’s function may be used to obtain a new
host Green’s function through the Dyson’s equation:
h i−1
G̃(ω) = Ĝ−1 + Σ̂ + ĤDC + ˆ − (µ − µ0 ) I . (34)
In general, the chemical potential, µ is found by fixing
the total occupancy from the full Green’s function, Ĝ to
be equal to the value found from DFT,
Z 0
1
− Im TrĜ = nDF
tot
T
, (35)
π −∞
where the trace is over spin and orbital indices.
Thus the full solution of the problem proceeds as fol-
lows. Given the ĤDF T (k), we guess an initial self-energy,
as well as the µ and µ0 ; and use these to find the local
and the host Green’s functions through Eqs. (33) and
(34). The host Green’s functions are then used to find
the self-energy, Σ̂ and Eqs.(33) and (35) are used to find
the chemical potential. For a fixed µ0 , these equations are
then iterated, until the self-energy converges. With the
and local Green’s functions. If the Luttinger’s theorem
is satisfied within a numerical tolerance, the solution is
considered to be obtained, else the µ0 is tuned, and the
DMFT equations are iterated, until the Luttinger’s the-
orem is satisfied.
The DFT predicted occupancy per spin on the three
correlated V-t2g orbitals in SrVO3 is 0.166, which implies
SrVO3 is a d1 system. For the DMFT calculations, we
employ interaction parameters U = 3.44 eV and J = 0.46
eV, that were obtained by Taranto et al.84 through the
random phase approximation (RPA). For SrVO3 , we have
not introduced explicit double counting correction be-
cause we choose the correlated subspace that is identi-
cal with the set of Wannier bands. We absorb the dou-
ble counting correction and orbital energies in the lattice
chemical potential, which we find by using Eq. (35).
Our computed GGA+DMFT spectrum for SrVO3 is

shown in Fig. 19 and compared with results obtained
from other impurity solvers. The GGA result (shown in
blue) has no signatures of correlation, while each of the
DMFT calculations exhibit a three peak structure. The
CTQMC results from GW+DMFT (black) agree quali-
tatively with those from LDA+DMFT. However, the de-
tails do differ. Namely, the positions and weights of the
resonance at the Fermi level and of the Hubbard bands
differ to a significant extent. This difference, naturally,
can be attributed to the different starting points, namely
GW vs LDA, of the CTQMC calculations. Results from
the MO-IPT solver agree with those from CTQMC in the
neighborhood of the chemical potential as well as in the
proximity of the lower Hubbard band. The upper Hub-
bard band is clearly in disagreement with the CTQMC
results.
FIG. 20. (color online) Comparison of photo emission
spectra obtained from different methods GW+DMFT84 ,
GGA+DMFT(MO-IPT), LDA+DMFT(CTQMC)84 and
experiment85 .
As a final benchmark of the GGA+DMFT(MO-IPT)
calculation, we compare our result with the exper-
imentally measured photo emission spectrum (PES)
which is shown in Fig. 20. A Hubbard satellite at
∼ −1.5 eV is seen in the experimental PES spec-
trum. Our GGA+DMFT(MO-IPT) calculation predicts
the Hubbard satellite at -1.25 eV. Results from other
approaches, namely LDA, LDA+DMFT(CTQMC) and
GW+DMFT(CTQMC) are also reproduced. Surpris-
ingly, the closest match with the experiment is achieved
by the GGA+DMFT(MO-IPT) in terms of the position
and width of the resonance at the Fermi level and of the
lower Hubbard band. Thus, we infer that the MO-IPT
method outlined in this work may be used as an efficient
16
tool to study the electronic structure of real material sys-
tems.
IV. Conclusions
The development of iterated perturbation theory
as an impurity solver for single band models and for
multi-band models dates back to almost two decades.
Although a few comparisons with numerically exact
methods have been made, being a perturbative ap-
proach, the method has suffered from reliability issues,
especially for multi-orbital systems. Nevertheless, sev-
eral multi-orbital extensions of IPT have been proposed
and used to investigate model Hamiltonians and even
real material systems. In this work, we have outlined a
multi-orbital extension of IPT, and benchmarked it ex-
tensively against continuous time quantum Monte Carlo
results. Our work is the first systematic study of the
multi-orbital Hubbard model using MO-IPT as a solver
and varying parameters such as filling, Hund’s coupling,
and Coulomb repulsion, as well as including crystal field
effects and application to real materials. One of the
main bottlenecks in methods based on spectral moment
expansions is the evaluation of high-order correlation
functions. We find that including such correlations that
are beyond two-particle type through approximate meth-
ods such as CPA or lower order decomposition, can lead
to spurious features at the chemical potential. We find
the best benchmarks simply by neglecting correlations
beyond two-particle. We conjecture that evaluation of
the higher-order correlations through exact methods
such as ligand field theory might be able to circumvent
the issues mentioned above86,87 . We are presently
implementing such a procedure. This procedure will
also enable us to treat the Hund’s coupling term in the
rotationally invariant form rather than the simpler and
approximate density-density type treated in the present
work. We are also planning to extend our method to
incorporate the off-diagonal hybridization, which at
present is very difficult to handle by numerical exact
methods like CTQMC. Apart from the benchmarks
for model Hamiltonians in various parameter regimes,
we have also carried out a GGA+DMFT(MO-IPT)
study of the perovskite SrVO3 , and compared our
predicted photoemission spectrum with experiments
and results from other methods. The agreement with
experiments was found to be excellent. A full scale
implementation of the method outlined here, with de-
tailed instructions for installation and use may be found
at http://www.institute.loni.org/lasigma/package/mo-
ipt/.
V. Acknowledgments
We thank CSIR and DST (India) for research funding.
Our simulations used an open source implementation88
 17
X
of the hybridization expansion continuous-time quan- α2 = −[(α − µ) + Uαβ hnβ i] , (40)
tum Monte Carlo algorithm89 and the ALPS40 libraries. β6=(α)
This work is supported by NSF DMR-1237565 and NSF
EPSCoR Cooperative Agreement EPS-1003897. Super-
computer support is provided by the Louisiana Opti- µαα αα αα 2
2 µ0 − (µ1 )
α3 = − , (41)
cal Network Initiative (LONI) and HPC@LSU. NSV ac- µαα αα
1 µ0
knowledges an international travel fellowship award from
IUSSTF. Dasari acknowledges the hospitality of the de-
partment of Physics & Astronomy and the Center for 1 X 2 X
µαα 2
2 = (α − µ) + Vkα + 2(α − µ) Uαβ hnβ i
Computation & Technology, at Louisiana State Univer- N
kα β6=α
sity. XX
+ Uαβ Uαγ hnβ nγ i , (42)
β6=α γ6=α

VI. Appendix
−1
V 2 kα −
P P
N kα β,γ6=α Uαβ Uαγ (hnβ nγ i − hnβ ihnγ i)
In this appendix, we provide the derivations of the un- α3 = P .
(α − µ) + β6=α Uαβ hnβ i
known parameters appearing in the MO-IPT ansatz for (43)
the self energy (Eq. (9)). For sufficiently large values of z, one can truncate the
Derivation for Aα : continued fraction expansion of the Green’s function
(Eq. (38)) at the appropriate level and take the limit
The spectral representation of the αth -orbital Green’s z → ∞. Up to the second order moment
function is given by
Z ∞ αα α1
D ()d Gαα (z) = . (44)
Gαα (z) = . (36) z + α2 − α2 α3
−∞ z− z

After substituting the continued fraction expansion coef-

This can be Taylor expanded to obtain the Green func-
ficients in Eq. (44), we find the self energy contribution
tion in terms of spectral moments,
to the Green’s function in the high frequency limit as
∞ ∞
Dαα ()d  2
Z X µn ω→∞
X
Gαα (z) = (1 + + 2 + · · · ) = n+1
, Σα (ω) −−−−→ Uαβ hnβ i
−∞ z z z n=0
z β6=α
(37) P
Uαβ Uαγ (hnβ nγ i − hnβ ihnγ i)
β,γ6=α
where µn ’s are the spectral moments. We can also repre- + ,
sent the Green function in terms of a continued fraction ω
expansion and this is given by
2
P
α1 X β6=α Uαβ hnβ i (1 − hnβ i)
Gαα
(z) = α2 . (38) Σα (ω) = Uαβ hnβ i +
z+ a ω
1+ z+a 3+··· β6=α
4 P P
β6=α γ6=(β6=α) Uαβ Uαγ (hnβ nγ i − hnβ ihnγ i)
By comparing Eq. (38) with Eq. (37), we obtain the con- + .
tinued fraction expansion coefficients in terms of spec- ω
(45)
tral moments. Now we can calculate the spectral mo-
ments exactly up to any order by using the following In the high frequency limit the self energy ansatz reduces
expressions30 : to the following form:
  X X (2)
Uαβ hnβ i + Aα
* +
Σα (ω) = Σαβ . (46)
αα †
µn = [...[[fα , H], H], ...H]− , [H, ...[H, [H, fα ]]...]− 
 
β6=(α) β6=(α)
| {z } | {z }
(n-p)-fold p-fold
+ It is easy to show that in the limit of high frequencies,
(2)
n = 0,1,2,.... ; 0≤ p ≤ n Σαβ has the following form33 ,
The relation between the first few spectral moments and 2
the continued fraction expansion coefficients is given by, (2) Uαβ
Σαβ = hn0β i (1 − hn0β i) . (47)
ω
α1 = µαα
0 = h{fα , fα† }i = 1, (39)
Here n0β is the Hartree-corrected impurity charge be-
cause the propagators used in the second order pair bub-
†
α2 = −µαα
1 = h{[fα , Himp ], fα }i , ble diagram are Hartree-corrected propagators. We ob-
 18

tain the expression for Aα by substituting Eq. (47) in µαα αα αα 2

1 µ3 − (µ2 )
and α4 = − αα αα αα . (53)
Eq. (46) and comparing with Eq. (45) as µ1 µ2 − (µ1 )2
2
P
β6=(α) Uαβ hnβ i (1 − hnβ i) In the atomic limit (V→ 0), the second order pair bubble
Aα = P (2)
2
β6=(α) Uαβ hn0β i (1 − hn0β i)
diagram Σαβ (ω) reduces to the following form22,28 ,
P P
β6=(α) Uαβ γ6=(β6=α) Uαγ (hnβ nγ i − hnβ ihnγ i) 2
Uαβ [hn0β i(1 − hn0β i)]
+ P 2 . (2)
Σαβ (ω) = . (54)
β6=(α) Uαβ hn0β i (1 − hn0β i) ω + + µ0
Note that a two-particle correlation function is needed to Here µ0 is the pseudo-chemical potential. As mentioned
find Aα . earlier, we find this quantity by satisfying the Luttinger’s
Derivation for Bα : theorem or equivalently the Friedel’s sum rule. Now, the
self energy ansatz becomes
The relation between the impurity Green’s function X
and the self energy in the atomic limit is, Σα = Uαβ hnβ i +
β6=(α)
1
Gα (ω) = + , (48) 2
P
Aα Uαβ [hn0β i(1 − hn0β i)]
ω + µ − α − Σα (ω) β6=(α)
2 [hn i(1 − hn i)](55)
.
P
ω + + µ0 − Bα β6=(α) Uαβ 0β 0β
where the self-energy, Σα (ω) may be represented as a
continued fraction: By comparing Eq. (55) with Eq. (49) we find the expres-
sion for Bα in terms of spectral moments as,
1
Σα (ω) = ω + + µ − α − α1 . (49)
z+
α2 µ0 − (α3 + α4 )
1+
a3 Bα = P 2 . (56)
β6=(α) Uαβ [hn0β i(1 − hn0β i)]
z+a4 ···

As a simple case we consider only two poles in the self

energy. In principle we can keep all the poles of the self After substituting the spectral moments in Eq. (56) Bα
energy but the difficulty is that a pole of order n in- becomes,
volves the (n + 1)th order correlation function. These P
µ0 + α − µ − β6=α Uαβ hnβ i
functions are very hard to calculate without making ap- Bα =
τ
proximations. With the two pole approximation for the P P P α
self energy, Eq. (49) reduces to the following form38 β6=α γ6=α η6=α Uαβ Uαγ Uαη [hnβ ihnγ nη i − hnβ nγ nη i]
− ,
X α2 α3 τα2 Aα
Σα (ω) = Uαβ hnβ i + (50) (57)
ω+ + α3 + α4
β6=(α)
where
X
2
where α2 = −µαα
1 , (51) τα = Uαβ hn0β i(1 − hn0β i). (58)
β6=α

µαα αα 2
2 − (µ1 )
α3 = − , (52)
µαα
1

∗
E-mail:nagamalleswararao.d@gmail.com
† 6
E-mail:raja@jncasr.ac.in
1
A. Georges, G. Kotliar, W. Krauth, and M. J. Rozenberg,
Rev. Mod. Phys. 68, 13 (Jan 1996), http://link.aps.
7
org/doi/10.1103/RevModPhys.68.13
2
E. Gull, Continuous-Time Quantum Monte Carlo Algo-
rithms for Fermions, Ph.D. thesis, ETH Zürich (December
8
2008)
3
M. Jarrell and J. E. Gubernatis, Physics Reports 269, 133
(Jan 1996)
4 9
R. Bulla, T. A. Costi, and T. Pruschke, Rev. Mod. Phys.
80, 395 (Apr 2008), http://link.aps.org/doi/10.1103/
RevModPhys.80.395
5 10
T. Pruschke and R. Bulla, Eur. Phys. J. B 44,
217 (April 2005), http://dx.doi.org/10.1140/epjb/
e2005-00117-4
M. Caffarel and W. Krauth, Phys. Rev. Lett. 72,
1545 (Mar 1994), http://link.aps.org/doi/10.1103/
PhysRevLett.72.1545
D. J. Garcı́a, K. Hallberg, and M. J. Rozenberg, Phys. Rev.
Lett. 93, 246403 (Dec 2004), http://link.aps.org/doi/
10.1103/PhysRevLett.93.246403
A. Rüegg, E. Gull, G. A. Fiete, and A. J. Millis, Phys. Rev.
B 87, 075124 (Feb 2013), http://link.aps.org/doi/10.
1103/PhysRevB.87.075124
T. A. Costi, J. Kroha, and P. Wölfle, Phys. Rev. B
53, 1850 (Jan 1996), http://link.aps.org/doi/10.1103/
PhysRevB.53.1850
K. Haule, S. Kirchner, J. Kroha, and P. Wölfle, Phys. Rev.
B 64, 155111 (Sep 2001), http://link.aps.org/doi/10.

1103/PhysRevB.64.155111
11
K. Morita, H. Maebashi, and K. Miyake, Physica B: Con-
densed Matter 312313, 547 (2002), ISSN 0921-4526,
the International Conference on Strongly Correlated Elec-
tron Systems, http://www.sciencedirect.com/science/
article/pii/S0921452601015320
12
G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko,
O. Parcollet, and C. A. Marianetti, Rev. Mod. Phys.
78, 865 (Aug 2006), http://link.aps.org/doi/10.1103/
RevModPhys.78.865
13
V. Drchal, V. Janis̆, J. Kudrnovský, V. S. Oudovenko,
X. Dai, K. Haule, and G. Kotliar, Journal of Physics: Con-
densed Matter 17, 61 (2005), http://stacks.iop.org/
0953-8984/17/i=1/a=007
14
D. E. Logan, M. P. Eastwood, and M. A. Tusch, Journal
of Physics: Condensed Matter 10, 2673 (1998), http://
stacks.iop.org/0953-8984/10/i=12/a=009
15
M. T. Glossop and D. E. Logan, Journal of Physics: Con-
densed Matter 14, 6737 (2002), http://stacks.iop.org/
0953-8984/14/i=26/a=313
16
M. R. Galpin, A. B. Gilbert, and D. E. Logan, Journal
of Physics: Condensed Matter 21, 375602 (2009), http:
//stacks.iop.org/0953-8984/21/i=37/a=375602
17
P. Kumar, Valence fluctuations and disorder effects in
strongly correlated electronic systems, Ph.D. thesis, Jawa-
harlal Nehru Centre for Advanced Scientific Research (De-
cember 2013)
18
A. Martı́n-Rodero, E. Louis, F. Flores, and C. Tejedor,
Phys. Rev. B 33, 1814 (Feb 1986), http://link.aps.org/
doi/10.1103/PhysRevB.33.1814
19
K. Yosida and K. Yamada, Progress of Theoreti-
cal Physics Supplement 46, 244 (1970), http://ptps.
oxfordjournals.org/content/46/244.abstract
20
A. Georges and G. Kotliar, Phys. Rev. B 45,
6479 (Mar 1992), http://link.aps.org/doi/10.1103/
PhysRevB.45.6479
21
H. Kajueter, Interpolative perturbation scheme for corre-
lated electron systems, Ph.D. thesis, Rutgers University,
New Brunswick (october 1996)
22
A. Martı́n-Rodero, F. Flores, M. Baldo, and R. Pucci,
Solid State Communications 44, 911 (1982),
http://www.sciencedirect.com/science/article/
pii/0038109882903039
23
E. Mller-Hartmann, Zeitschrift fr Physik B Condensed
Matter 76, 211 (1989), ISSN 0722-3277, http://dx.doi.
org/10.1007/BF01312686
24
K. Yamada, Progress of Theoretical Physics 53, 970
(1975), http://ptp.oxfordjournals.org/content/53/4/
970.abstract
25
X. Y. Zhang, M. J. Rozenberg, and G. Kotliar, Phys. Rev.
Lett. 70, 1666 (Mar 1993), http://link.aps.org/doi/
10.1103/PhysRevLett.70.1666
26
T. Schäfer, G. Rohringer, O. Gunnarsson, S. Ciuchi,
G. Sangiovanni, and A. Toschi, Phys. Rev. Lett. 110,
246405 (Jun 2013), http://link.aps.org/doi/10.1103/
PhysRevLett.110.246405
27
E. Kozik, M. Ferrero, and A. Georges, Phys. Rev. Lett.
114, 156402 (Apr 2015), http://link.aps.org/doi/10.
1103/PhysRevLett.114.156402
28
H. Kajueter and G. Kotliar, Phys. Rev. Lett. 77,
131 (Jul 1996), http://link.aps.org/doi/10.1103/
PhysRevLett.77.131
29
H. Kajueter, G. Kotliar, and G. Moeller, Phys. Rev. B 53,
16214 (Jun 1996), http://link.aps.org/doi/10.1103/
19
PhysRevB.53.16214
30
M. Potthoff, T. Wegner, and W. Nolting, Phys. Rev. B 55,
16132 (Jun 1997), http://link.aps.org/doi/10.1103/
PhysRevB.55.16132
31
T. Wegner, M. Potthoff, and W. Nolting, Phys. Rev.
B 57, 6211 (Mar 1998), http://link.aps.org/doi/10.
1103/PhysRevB.57.6211
32
L.-F. Arsenault, P. Sémon, and A.-M. S. Tremblay, Phys.
Rev. B 86, 085133 (Aug 2012), http://link.aps.org/
doi/10.1103/PhysRevB.86.085133
33
A. L. Yeyati, F. Flores, and A. Martı́n-Rodero, Phys. Rev.
Lett. 83, 600 (Jul 1999), http://link.aps.org/doi/10.
1103/PhysRevLett.83.600
34
A. Liebsch, Phys. Rev. B 70, 165103 (Oct 2004), http:
//link.aps.org/doi/10.1103/PhysRevB.70.165103
35
M. S. Laad, L. Craco, and E. Müller-Hartmann, Phys. Rev.
Lett. 91, 156402 (Oct 2003), http://link.aps.org/doi/
10.1103/PhysRevLett.91.156402
36
T. Fujiwara, S. Yamamoto, and Y. Ishii, Journal of
the Physical Society of Japan 72, 777 (2003), http://
journals.jps.jp/doi/abs/10.1143/JPSJ.72.777
37
S. Y. Savrasov, V. Oudovenko, K. Haule, D. Villani, and
G. Kotliar, Phys. Rev. B 71, 115117 (Mar 2005), http:
//link.aps.org/doi/10.1103/PhysRevB.71.115117
38
V. Oudovenko, K. Haule, S. Y. Savrasov, D. Villani, and
G. Kotliar, “Interpolative approach for solving quantum
impurity model based on the slave–boson mean–field ap-
proximation,” Unpublished
39
E. Gull, A. J. Millis, A. I. Lichtenstein, A. N. Rubtsov,
M. Troyer, and P. Werner, Rev. Mod. Phys. 83, 349 (May
2011), http://link.aps.org/doi/10.1103/RevModPhys.
83.349
40
B. Bauer, L. D. Carr, H. G. Evertz, A. Feiguin, J. Freire,
S. Fuchs, L. Gamper, J. Gukelberger, E. Gull, S. Guertler,
A. Hehn, R. Igarashi, S. V. Isakov, D. Koop, P. N. Ma,
P. Mates, H. Matsuo, O. Parcollet, G. Pawowski, J. D.
Picon, L. Pollet, E. Santos, V. W. Scarola, U. Scholl-
wck, C. Silva, B. Surer, S. Todo, S. Trebst, M. Troyer,
M. L. Wall, P. Werner, and S. Wessel, Journal of Statis-
tical Mechanics: Theory and Experiment 2011, P05001
(2011), http://stacks.iop.org/1742-5468/2011/i=05/
a=P05001
41
M. Sentef, J. Kuneš, P. Werner, and A. P. Kampf, Phys.
Rev. B 80, 155116 (Oct 2009), http://link.aps.org/
doi/10.1103/PhysRevB.80.155116
42
L. de’ Medici, J. Mravlje, and A. Georges, Phys. Rev. Lett.
107, 256401 (Dec 2011), http://link.aps.org/doi/10.
1103/PhysRevLett.107.256401
43
M. Jarrell and J. E. Gubernatis, Physics Reports 269, 133
(Jan 1996)
44
H. Barman, Diagrammatic perturbation theory based in-
vestigations of the Mott transition physics, Ph.D. thesis,
Jawaharlal Nehru Centre for Advanced Scientific Research
(March 2011)
45
A. L. Fetter and J. D. Walecka, Quantum Theory Of Many-
Particle Systems (Dover Publications, 2003)
46
H. Terletska, J. Vučičević, D. Tanasković, and V. Do-
brosavljević, Phys. Rev. Lett. 107, 026401 (Jul 2011),
http://link.aps.org/doi/10.1103/PhysRevLett.107.
026401
47
R. Bulla, Phys. Rev. Lett. 83, 136 (Jul 1999), http://
link.aps.org/doi/10.1103/PhysRevLett.83.136
48
R. Bulla, T. A. Costi, and D. Vollhardt, Phys. Rev. B 64,
045103 (Jun 2001), http://link.aps.org/doi/10.1103/

PhysRevB.64.045103
49
K. Inaba, A. Koga, S.-i. Suga, and N. Kawakami, Phys.
Rev. B 72, 085112 (Aug 2005), http://link.aps.org/
doi/10.1103/PhysRevB.72.085112
50
P. Zhang, P. Reis, K.-M. Tam, M. Jarrell, J. Moreno,
F. Assaad, and A. K. McMahan, Phys. Rev. B 87,
121102 (Mar 2013), http://link.aps.org/doi/10.1103/
PhysRevB.87.121102
51
P. A. Lee, N. Nagaosa, and X.-G. Wen, Rev. Mod. Phys.
78, 17 (Jan 2006), http://link.aps.org/doi/10.1103/
RevModPhys.78.17
52
M. Z. Hasan and C. L. Kane, Rev. Mod. Phys. 82,
3045 (Nov 2010), http://link.aps.org/doi/10.1103/
RevModPhys.82.3045
53
J. Kuneš and V. I. Anisimov, Phys. Rev. B 78,
033109 (Jul 2008), http://link.aps.org/doi/10.1103/
PhysRevB.78.033109
54
V. Jaccarino, G. K. Wertheim, J. H. Wernick, L. R. Walker,
and S. Arajs, Phys. Rev. 160, 476 (Aug 1967), http://
link.aps.org/doi/10.1103/PhysRev.160.476
55
C. Petrovic, Y. Lee, T. Vogt, N. D. Lazarov,
S. L. Bud’ko, and P. C. Canfield, Phys. Rev. B 72,
045103 (Jul 2005), http://link.aps.org/doi/10.1103/
PhysRevB.72.045103
56
A. Garg, H. R. Krishnamurthy, and M. Randeria, Phys.
Rev. Lett. 97, 046403 (Jul 2006), http://link.aps.org/
doi/10.1103/PhysRevLett.97.046403
57
A. J. Kim, M. Y. Choi, and G. S. Jeon, Phys. Rev. B 89,
165117 (Apr 2014), http://link.aps.org/doi/10.1103/
PhysRevB.89.165117
58
L. Tsui, H.-C. Jiang, Y.-M. Lu, and D.-H. Lee, Nu-
clear Physics B 896, 330 (2015), ISSN 0550-3213,
http://www.sciencedirect.com/science/article/pii/
S0550321315001534
59
A. Amaricci, J. C. Budich, M. Capone, B. Trauzettel, and
G. Sangiovanni, Phys. Rev. Lett. 114, 185701 (May 2015),
http://link.aps.org/doi/10.1103/PhysRevLett.114.
185701
60
A. E. Antipov, I. S. Krivenko, V. I. Anisimov, A. I.
Lichtenstein, and A. N. Rubtsov, Phys. Rev. B 86,
155107 (Oct 2012), http://link.aps.org/doi/10.1103/
PhysRevB.86.155107
61
S. Biermann, L. de’ Medici, and A. Georges, Phys. Rev.
Lett. 95, 206401 (Nov 2005), http://link.aps.org/doi/
10.1103/PhysRevLett.95.206401
62
L. de’ Medici, Phys. Rev. B 83, 205112 (May 2011), http:
//link.aps.org/doi/10.1103/PhysRevB.83.205112
63
G. Kotliar, S. Y. Savrasov, K. Haule, V. S. Oudovenko,
O. Parcollet, and C. A. Marianetti, Rev. Mod. Phys.
78, 865 (Aug 2006), http://link.aps.org/doi/10.1103/
RevModPhys.78.865
64
P. Werner and A. J. Millis, Phys. Rev. Lett. 99,
126405 (Sep 2007), http://link.aps.org/doi/10.1103/
PhysRevLett.99.126405
65
E. Pavarini, S. Biermann, A. Poteryaev, A. I. Lichtenstein,
A. Georges, and O. K. Andersen, Phys. Rev. Lett. 92,
176403 (Apr 2004), http://link.aps.org/doi/10.1103/
PhysRevLett.92.176403
66
E. Pavarini, A. Yamasaki, J. Nuss, and O. K. Andersen,
New Journal of Physics 7, 188 (2005), http://stacks.
iop.org/1367-2630/7/i=1/a=188
67
A. I. Poteryaev, M. Ferrero, A. Georges, and O. Parcollet,
Phys. Rev. B 78, 045115 (Jul 2008), http://link.aps.
org/doi/10.1103/PhysRevB.78.045115
20
68
N. Parragh, G. Sangiovanni, P. Hansmann, S. Hum-
mel, K. Held, and A. Toschi, Phys. Rev. B 88,
195116 (Nov 2013), http://link.aps.org/doi/10.1103/
PhysRevB.88.195116
69
K. Held, Advances in Physics 56, 829 (OCT 2007)
70
O. Andersen, A. Liechtenstein, O. Jepsen, and F. Paulsen,
Journal of Physics and Chemistry of Solids 56,
1573 (1995), ISSN 0022-3697, Proceedings of the
Conference on Spectroscopies in Novel Superconduc-
tors, http://www.sciencedirect.com/science/article/
pii/0022369795002693
71
N. Marzari and D. Vanderbilt, Phys. Rev. B 56 (Nov
1997), doi:“bibinfo doi 10.1103/PhysRevB.56.12847, http:
//link.aps.org/doi/10.1103/PhysRevB.56.12847
72
H. T. Dang, X. Ai, A. J. Millis, and C. A. Marianetti,
Phys. Rev. B 90, 125114 (Sep 2014), http://link.aps.
org/doi/10.1103/PhysRevB.90.125114
73
P. Blaha, K. Schwarz, G. Madsen, D. Kvasnicka, and
J. Luitz, Computer Code WIEN2k: An Augmented Plane
Wave plus Local Orbitals Program for Calculating Crys-
tal Properties (Vienna University of Technology, Vienna,
2001)
74
M. Rey, P. Dehaudt, J. Joubert, B. Lambert-Andron,
M. Cyrot, and F. Cyrot-Lackmann, Journal of Solid
State Chemistry 86, 101 (1990), ISSN 0022-4596,
http://www.sciencedirect.com/science/article/pii/
002245969090119I
75
G. Trimarchi, I. Leonov, N. Binggeli, D. Korotin, and
V. I. Anisimov, Journal of Physics: Condensed Matter
20, 135227 (2008), http://stacks.iop.org/0953-8984/
20/i=13/a=135227
76
J. P. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev.
Lett. 77, 3865 (Oct 1996), http://link.aps.org/doi/10.
1103/PhysRevLett.77.3865
77
K. Jan, R. Arita, P. Wissgott, A. Toschi, H. Ikedaa, and
K. Held, Comp. Phys. Commun. 181, 1888 (2010)
78
A. A. Mostofi, J. R. Yates, Y.-S. Lee, I. Souza, D. Van-
derbilt, and N. Marzari, “Wannier90: A Tool for Ob-
taining Maximally-Localised Wannier Functions,” Comp.
Phys. Commun. 178, 685 (2008)
79
H. Ishida, D. Wortmann, and A. Liebsch, Phys. Rev. B 73,
245421 (Jun 2006), http://link.aps.org/doi/10.1103/
PhysRevB.73.245421
80
I. A. Nekrasov, G. Keller, D. E. Kondakov, A. V.
Kozhevnikov, T. Pruschke, K. Held, D. Vollhardt, and
V. I. Anisimov, Phys. Rev. B 72, 155106 (Oct 2005), http:
//link.aps.org/doi/10.1103/PhysRevB.72.155106
81
F. Lechermann, A. Georges, A. Poteryaev, S. Biermann,
M. Posternak, A. Yamasaki, and O. K. Andersen, Phys.
Rev. B 74, 125120 (Sep 2006), http://link.aps.org/
doi/10.1103/PhysRevB.74.125120
82
N. Parragh, A. Toschi, K. Held, and G. Sangiovanni, Phys.
Rev. B 86, 155158 (Oct 2012), http://link.aps.org/
doi/10.1103/PhysRevB.86.155158
83
N. Parragh, Strongly Correlated Multi-Orbital Systems A
Continuous-Time Quantum Monte Carlo Analysis, Ph.D.
thesis, Universität Würzburg (December 2013)
84
C. Taranto, M. Kaltak, N. Parragh, G. Sangiovanni,
G. Kresse, A. Toschi, and K. Held, Phys. Rev. B 88,
165119 (Oct 2013), http://link.aps.org/doi/10.1103/
PhysRevB.88.165119
85
A. Sekiyama, H. Fujiwara, S. Imada, S. Suga, H. Eisaki,
S. Uchida, K. Takegahara, H. Harima, Y. Saitoh,
I. Nekrasov, G. Keller, D. Kondakov, A. Kozhevnikov,

T. Pruschke, K. Held, D. Vollhardt, and V. Anisimov,
Phys. Rev. Lett. 93, 156402 (Oct 2004), http://link.
88
aps.org/doi/10.1103/PhysRevLett.93.156402
86
O. Miura and T. Fujiwara, Phys. Rev. B 77,
89
195124 (May 2008), http://link.aps.org/doi/10.1103/
PhysRevB.77.195124
87
A. I. Lichtenstein and M. I. Katsnelson, Phys. Rev.
21
B 57, 6884 (Mar 1998), http://link.aps.org/doi/10.
1103/PhysRevB.57.6884
H. H. Hafermann, P. Werner, and E. Gull, Computer
Physics Communications, 184, 1280 (2013)
P. Werner, A. Comanac, L. de’ Medici, M. Troyer, and
A. J. Millis, Phys. Rev. Lett. 97, 076405 (Aug 2006), http:
//link.aps.org/doi/10.1103/PhysRevLett.97.076405